inspecting-hermes-desktop-dom
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local shell commands and scripts to interact with the development environment.
- It uses
curlto verify the status of the CDP port on the local loopback address127.0.0.1. - It executes local JavaScript files via
node apps/desktop/scripts/eval.mjsto perform UI queries. - It launches isolated instances of the application using
npx electron .with specific user data directories. - [PROMPT_INJECTION]: The skill facilitates reading live application content, which introduces a vector for indirect prompt injection if the application displays untrusted data.
- Ingestion points: Data is retrieved from the browser context via the
cdp.evalmethod and thescripts/eval.mjsscript as described in SKILL.md. - Boundary markers: None detected. The instructions do not specify the use of delimiters or 'ignore' instructions when processing the content of the DOM.
- Capability inventory: The agent has the capability to execute shell commands (
node,npx,curl) and access the local file system (via--user-data-dirandHERMES_HOMEenvironment variables). - Sanitization: No explicit sanitization or filtering of the DOM content is described before the agent processes it.
Audit Metadata